industry growth through spinoffs and startupsrieti discussion paper series 17-e-057 march 2017...
TRANSCRIPT
DPRIETI Discussion Paper Series 17-E-057
Industry Growth through Spinoffs and Startups
OHYAMA AtsushiHitotsubashi University
The Research Institute of Economy, Trade and Industryhttp://www.rieti.go.jp/en/
RIETI Discussion Paper Series 17-E-057 March 2017
Industry Growth through Spinoffs and Startups*
OHYAMA Atsushi
Hitotsubashi University
Abstract
The literature on industry life cycle suggests that there is some underlying mechanism that generates differences
in time for industries reaching their peaks. What causes variation in such peak times across industries? In this
paper, I use the Japanese Census of Manufacture and investigate (i) whether creation and destruction of
submarkets in an industry affect the length of positive net entry periods and subsequent entry rates, (ii) what
type of firm is more likely to be actively engaged in a newly created or destructed submarket, and (iii) how
reallocation of unrealized opportunities from incumbent firms to spinoff firms affects the entry process. This study
reveals that the creation and destruction of a submarket allow an industry to continue attracting new entrants,
that startup and spinoff firms are more likely to enter a newly created submarket than incumbent firms, and that
new entry is encouraged when unrealized business opportunities are reallocated smoothly.
Keywords: Industry growth, Product innovation, Industry life cycle, Submarket JEL classification: O31, O47
RIETI Discussion Papers Series aims at widely disseminating research results in the form of professional papers, thereby stimulating lively discussion. The views expressed in the papers are solely those of the author(s), and neither represent those of the organization to which the author(s) belong(s) nor the Research Institute of Economy, Trade and Industry.
*This study is conducted as a part of the βMicroeconometric Analysis of Firm Growthβ project undertaken at the Research Institute of Economy, Trade and Industry (RIETI). This study utilizes the micro data of the questionnaire information based on the βBasic Survey of Japanese Business Structure and Activities,β βCensus of Manufacture,β and βEconomic Census for Business Activity (H24)β which are conducted by the Ministry of Economy, Trade and Industry (METI), and the Kikatsu Oyako converter, which is provided by RIETI. The author is grateful for helpful comments and suggestions by Discussion Paper seminar participants at RIETI.
1
1. Introduction
Some industries continue to grow and attract new entrants for a long period of time and other
industries stop doing so within a very short period of time. Klepper (2016), for example, document
that it took 15 years or more for the number of firms to reach their peaks in the U.S. tire and
automobile industries whereas the number of firms in the penicillin and TV receiver industries
reached their peaks very quickly by taking only less than 10 years. The literature on the industry
life-cycle has documented a stylized fact that many industries start off with a very few number of
firms, but the number of active firms rapidly increases through new entry until a shakeout takes
place and, after the shakeout, the industries are typically dominated by a few firms. (Gort and
Klepper, 1982; Klepper and Graddy, 1990; Filson, 2001). Since many industries experience this
life-cycle pattern, there seems some underlying mechanism that generates differences in the length
of periods for an industry reaching its peaks (i.e., shakeout).
What causes such differences across industries? In this paper, I try to answer this question
empirically by focusing on a relationship between new entry and the evolution of submarkets (i.e.,
product innovation process) in a given industry as well as roles of firm heterogeneity played out in
product innovation and industry growth. More specifically, I ask a set of the following three
questions to investigate a mechanism that encourages or discourages new entry in a given industry.
The first question is whether creation and destruction of submarkets in an industry affect the length
of positive net entry periods and subsequent entry rates in that industry. The second question asks
what type of firms β startup firms, spinoff firms, or incumbent firms β are more likely to be actively
engaged in a newly created or destructed submarket. In the third question, I ask how frictions to
the pursuit of business opportunities by incumbent firms and frictions to reallocation of unrealized
2
opportunities from incumbent firms to spinoff firms affect entry process at the industry level.
Although these three questions seem unrelated to one another, they are actually connected in the
following sense. There is anecdotal evidence that spinoff firms tend to pursue a business
opportunity their parent firm gave up pursuing and they then create a submarket and compete with
their parent firms by providing a new product in the industry where their parent firms currently
operate.
The main finding of this study regarding the first question is that the creation and
destruction of a submarket allow an industry to continue attracting new entrants so that the timing
of a shakeout is delayed. This result is partially consistent with Klepperβs (2016) conjecture that an
industry continues to attract new entrants when a submarket is created within that industry. In
addition to this conjecture, this study reveals the importance of submarket destruction for industry
dynamics as well. This study also finds that establishments belonging to startup firms or spinoff
firms are more likely to enter a newly created market than establishments belonging to incumbent
firms. This indicates that startup firms and spinoff firms are the main player in a new submarket.
Regarding the third question, this study shows that new entry is encouraged when unrealized
business opportunities are reallocated smoothly to spinoff firms from incumbent firms. This finding
can be interpreted as indicating that reallocation of unrealized opportunities from incumbent firms
to spinoff firms sparks subsequent entries. The cumulative nature of business and innovation
opportunities generates self-enhancing process through which realization of one opportunity
becomes a basis for the realization of a next opportunity. In this light, smooth reallocation of
unrealized opportunities allows firms to offer new but similar products in an industry (i.e., creation
3
of submarket) and the industry results in continuing to attract new entrants.
The main contribution of this paper is three-fold. Shakeout phenomenon itself has
attracted a lot of attention, and several mechanisms that cause a shakeout have been proposed in
the industry evolution literature (Jovanovic and MacDonald, 1994; Klepper and Simons, 2000;
Barbarino and Jovanovic, 2007), but we have known a very little about why time to reach a peak
of the number of active firms differs from industry to industry. This study contributes to this line
of research by focusing on product innovation and firm heterogeneity and empirically examining
underlying forces for the entry process in industry evolution. The second contribution of this study
is to generalize anecdotal evidence and case studies claiming that creation and destruction of
products influence a course of industry dynamics (i.e., Schumpeterian industry growth) and that
submarket creation is positively associated with a length of industryβs prosperity (Geroski 1995;
Klepper and Thompson, 2006; Klepper and Golman, 2016). In this study, I use a large-scale data
set that contains more than 400 industries to test the generalizability of these claims. This
generalization is quite important because it provides sound policy implications for making a fast
and persistently growing industry. The third contribution of this study is to provide a new
perspective on reallocation effects on industry growth. Most extant studies focus on effects of
resource reallocation (Hsieh and Klenow, 2009) or technology reallocation (Collard-Wexler and
De Loecker, 2015) on industry and economic growth. This study tries to draw attention to
reallocation of unrealized business opportunities and examine its relationship to industry growth.
This study reveals that smooth reallocation of unrealized opportunities among different types of
firms is positively associated with industry growth in terms of the number of active firms.
4
The rest of this paper is organized as follows. Section 2 briefly reviews the literatures on
industry dynamics, spinoffs, real authority and resource reallocation. Section 3 outlines a
theoretical model to guide empirical analysis and interpretation of empirical results. The detailed
information of data used in this study is provided in Section 4. Section 5 is devoted to regression
analyses regarding submarket creation and destruction, entry and reallocation of unrealized
opportunities. Section 6 concludes.
2. Brief Literature Review
This research is related to four strands of the following literatures; Schumpeterian industry
growth, submarkets and spinoffs, formal and real authority, and resource reallocation. Since
Schumpeterβs (1942) work on innovation, economic development and industry dynamics have
been examined extensively through a lens of βcreative destruction.β What we call Schumpeterian
growth model was rigorously formalized by Aghion and Howitt (1992) to capture the process of
creative destruction in which old firms are replaced by new firms as well as to demonstrate that
this process is an important element of economic growth. This Schumpeterian growth model has
been also taken to the industry level and helped to improve our understanding of industry dynamics.
In particular, the formal model by Klette and Kortum (2004) explicitly incorporates the creation of
new goods and the distinction of old goods into an analytical framework and shows that industry
dynamics is driven by the process of creative destruction. Their model is quite successful for
explaining many stylized facts from firm-level studies about firm growth, entry, exit and size
distribution. Lentz and Mortensen (2007) extend the model of Klette and Kortum (2004) to explain
5
a link between growth and firm productivity and effects of reallocation of resources on the
aggregate growth. They use Danish data and show that continual reallocation of resources towards
new and growing firms through the process of creative destruction accounts for three quarters of
the growth in the modeled economy. Regarding the relationship between new market creation and
industry growth, Tang (2016) observes that the Japanese manufacturing sector grew through the
appearance of new sectors during the period through 1868 to1912.
We can observe a substantial amount of variation in heterogeneity of firm activity within
an industry if one defines the industry by the SIC code. By drawing attention to this heterogeneity,
Klepper and Thompson (2006) propose that an industry consists of various βsubmarketsβ and argue
that regularities about firm growth, entry and exist can be explained well if we take the creation
and destruction process of submarkets into account. They take their theoretical predictions to the
data about US laser industry where several types of lasers were produced between 1961 and 1994.
They found supporting evidence such as that the number of lasers produced by all firms increase
with age and that the probability of exit is a decreasing function of the number of lasers produced.
In relation to these submarket phenomena, Klepper and Sleeper (2004) present the evidence that
spinoff firms and spinout firms tend to produce similar types of products as their parent firms and
they therefore paly an important role in the creation of submarkets.
One may argue that spinoff firms are heavily influenced or actually controlled by their
former parent firms in spite of the fact that they are a legally independent entity. In particular, this
argument appears to fit to Japanese firms better than firms in western countries. Based on the
concept of formal and real authority proposed by Aghion and Tirole (1997), Itoh and Hayashida
6
(1997) theoretically demonstrate that the problem of over-intervention can be mitigated and
efficient outcomes about human capital investments can be achieved when authority is delegated
to spinoff firms from parent firms. Otsubo (2005) uses a sample of 300 large Japanese firms in the
Japanese Company Handbook presents the indirect evidence that the main function of spinoff in
cases of Japanese firms is to give workers incentives to make relation-specific investments.
It has been widely recognized that resource allocation is an important channel for industry
dynamics through which firmβs entry, exit and productivity are influenced significantly. Hseih and
Klenow (2009) discuss that distortions in resource allocation are largely responsible for a low level
of productivity and show that the aggregate productivity of China and India would increase by
about 30 to 50 percent in China and 40 to 60 percent in India if their resources were reallocated to
the efficiency level of the United States. Since their seminal work, similar findings about
misallocation have been reported by using different data sets around the world. Rather than
resource allocation itself, Collard-Wexler and De Loecker (2015) focus on the allocation of
technologies as a source of productivity growth. They point to the importance of allocation of
technology by showing that the replacement of old production technology by new technology
explains the half of the aggregate productivity growth in the US steel industry and that resources
are allocated to more productivity firms during the period of technical changes.
3. Conceptual Framework
In this section, a conceptual framework is presented to serve as guiding empirical analyses in
Section 5. I extend the industry growth model of Klette and Kortum (2004) and that of Lentz and
7
Mortensen (2007) by incorporating firm heterogeneity and frictions to innovation opportunities
into the analytical framework.
3. 1. Demand and Prices
Time flows continuously and inter-temporal utility of the representative household at time t is given
by
πππ‘π‘ = οΏ½ ππβππ(π π βπ‘π‘)πππππΆπΆπ‘π‘β
π‘π‘ππππ
where πΆπΆπ‘π‘ is the amount of aggregate consumption, and ππ is the interest rate. There are different
products and the aggregate consumption is given by the CES function:
πΆπΆπ‘π‘ = οΏ½οΏ½ οΏ½π΄π΄π‘π‘(ππ)π₯π₯π‘π‘(ππ)οΏ½ππβ1ππ
1
0πππποΏ½
ππππβ1
where π₯π₯π‘π‘(ππ) is the quantity of differentiated good j at time t and π΄π΄π‘π‘(ππ) is the quality of
differentiated good j at time t. The quality π΄π΄π‘π‘(ππ) depends on the number of successful innovations
in the past π½π½π‘π‘(ππ) and the size of improvement ππ:
π΄π΄π‘π‘(ππ) = π½π½π‘π‘(ππ)ππ
The profit maximization condition for the final good producer implies
πππ‘π‘(ππ)π΄π΄π‘π‘(ππ)π₯π₯π‘π‘(ππ) = οΏ½ πππ‘π‘πππ‘π‘(ππ)οΏ½
ππβ1ππ (1)
where πππ‘π‘(ππ) is the price of differentiated good j without adjusting the quality π΄π΄π‘π‘(ππ), πππ‘π‘ is the
price of final good, and ππ is the expenditure of the final good. It follows from the zero profit
condition of the final good producer and equation (1) that
πππ‘π‘πΆπΆπ‘π‘ = οΏ½ οΏ½πππ‘π‘πππ‘π‘(ππ)
οΏ½ππβ1
πππ‘π‘πΆπΆπ‘π‘1
0ππππ
8
which implies
πππ‘π‘ = οΏ½οΏ½ πππ‘π‘(ππ)1βππ1
0πππποΏ½
11βππ
Using the formula of quality adjusted price that πππ‘π‘(ππ) = π΄π΄π‘π‘(ππ)πππ‘π‘(ππ), we obtain
πππ‘π‘ = οΏ½οΏ½ οΏ½πππ‘π‘(ππ)π΄π΄π‘π‘(ππ)
οΏ½1βππ1
0πππποΏ½
11βππ
3. 2. Growth Rates and Innovation
Because of the assumption that πππ‘π‘πΆπΆπ‘π‘ = ππ, we have
πΆπΆπ‘π‘ =πππππ‘π‘
= ππ οΏ½οΏ½ οΏ½π΄π΄π‘π‘(ππ)πππ‘π‘(ππ)
οΏ½ππβ11
0πππποΏ½
1ππβ1
We assume that firms in each sector of the differentiated products are involved in a Bertrand
competition. As a result, the price charged by each firm is the limit price and its markup is given
simply by ππππβ1
. Denoting the probability of a successful innovation by πΏπΏ, the rate of growth in
the aggregate consumption is given by
ππ = πΏπΏ οΏ½ππππβ1
ππβ1οΏ½ (2)
3. 3. Incentives for Innovation
Consider an incumbent firm which produces n variety of goods and generates profit ππ from each
variety. The incumbent firm has an opportunity to make an R&D investment and create a new
variety. The R&D expenditure is given by πΆπΆ(πΌπΌ,ππ) , where πΌπΌ is the R&D intensity and πΆπΆ(β)
9
exhibits constant returns-to-scale. The incumbent firm may not pursue all the innovation
opportunities, and they may pursue πΌπΌ of the innovation opportunities.
The Bellman equation for the incumbent firm is given by
ππππ(ππ) = πππππ₯π₯πΌπΌ οΏ½ππππ β ππππ οΏ½πΌπΌπποΏ½ + πΌπΌ[ππ(ππ + 1) β ππ(ππ)] + (1 β πΌπΌ)ππ(ππ)οΏ½ (3)
where ππ is the value function of the incumbent firm, and ππ is the interest rate.
The first order condition and envelope condition imply
ππ(ππ) = π£π£ππ and
πΌπΌ(ππ) = ππππ where ππ and π£π£ are determined by
ππβ²(ππ) = πΌπΌπ£π£ and
[ππ β πΌπΌππ β (1 β πΌπΌ)]π£π£ = ππ β ππ(ππ)
It follows from these equations that ππππππππ
> 0 and ππππππππ
> 0. Thus, the per-product research
intensity ππ decreases as incumbent firms miss more innovation opportunities. Also, the per-
product profit increases the per-product research intensity ππ.
A spinoff firm is assumed to mainly pursue an innovation opportunity the incumbent firm
gave up pursuing. In other words, the spinoff firm faces 1 β πΌπΌ of the innovation opportunities. In
addition, there is (1 β πΎπΎ) degree of industry-level frictions to reallocation of the innovation
opportunities from incumbent firms to spinoff firms. Under the conditions, the Bellman equation
for the spinoff firm is given by
ππππ(ππ) = πππππ₯π₯πΌπΌ οΏ½ππππ β ππππ οΏ½πΌπΌπποΏ½ + (1 β πΌπΌ)πΎπΎ[ππ(ππ + 1) β ππ(ππ)] + πΌπΌ(1 β πΎπΎ)ππ(ππ)οΏ½ (4)
where ππ is the value function of the spinoff firm.
The first order condition and envelope condition yield
ππ(ππ) = π£π£ππ
10
and
πΌπΌ(ππ) = ππππ
where ππ and π£π£ are determined by
ππβ²(ππ) = (1 β πΌπΌ)πΎπΎπ£π£ and
[ππ β (1 β πΌπΌ)πΎπΎππ β πΌπΌ(1 β πΎπΎ)]π£π£ = ππ β ππ(ππ) A startup firm starts its business with one variety. Therefore, the Bellman equation for
the startup firm is given by Equation (3) with ππ = 1.
3. 4. Industry Dynamics
Let ππππ(π‘π‘) denote the measure of firms with n products in an industry at date t. As in Klette and
Kortum (2004), ππππ(π‘π‘) changes over time according to
οΏ½ΜοΏ½πππ(π‘π‘) = (ππ β 1)πΌπΌππππππβ1(π‘π‘) + (ππ + 1)πΏπΏππππ+1(π‘π‘) β ππ(πΌπΌππ + πΏπΏ)ππππ(π‘π‘)
for ππ β₯ 2 and
οΏ½ΜοΏ½π1(π‘π‘) = ππ + (1 β πΌπΌ)πΎπΎπΎπΎ + 2πΏπΏππ2(π‘π‘) β (πΌπΌππ + πΏπΏ)ππ1(π‘π‘)
Note that πΏπΏ = πΌπΌππ + (1 β πΌπΌ)πΎπΎπΎπΎ + ππ . When ππ > 0 , πΎπΎ > 0 and ππ > 0 , we have the
steady state mass of the firms as
ππππ =(πΌπΌππ)ππβ1(ππ + (1 β πΌπΌ)πΎπΎπΎπΎ)
πππΏπΏππ=π π πποΏ½
11 + π π
οΏ½ππ
where π π = ππ+(1βππ)πΎπΎπΎπΎππππ
. The steady state total mass of firms in a given industry is given by
ππ = π π ππππ οΏ½1+π π π π οΏ½ (5)
Equation (5) shows that the total mass of firms depend on (i) incentives to innovate by
different types of firms (i.e., ππ, πΎπΎ and ππ), and (ii) frictions to innovation opportunities (i.e., πΌπΌ
and πΎπΎ). The function π π increases with ππ and πΎπΎ and decreases with ππ. Since the total mass of
11
firms ππ is an increasing function of π π , the total mass of firms in an industry is larger when
innovation intensities of new entrants and spinoff firms are relatively higher than that of incumbent
firms. The function π π decreases with πΌπΌ and increases with πΎπΎ. Thus, the total mass of firms in an
industry is larger when incumbents miss innovation opportunities at many times so that spinoff
firms can utilize these opportunities to start a new business. Further, the total mass of firms in an
industry is larger when spinoff firms face a lesser degree of frictions (i.e., larger value of πΎπΎ) to start
a new business. In the reality, these frictions may include a non-competing clause, regulations,
taxes and so forth.
We can also examine the evolution of the number of firms in an industry. Note that
ππππ(π‘π‘) β π π ππ
[ππ(π‘π‘)]ππ
where
ππ(π‘π‘) =πΌπΌπππΏπΏοΏ½πΏπΏβπΏπΏππβοΏ½(1βππ)πΎπΎπΎπΎ+πποΏ½π‘π‘
πΏπΏβππππβοΏ½(1βππ)πΎπΎπΎπΎ+πποΏ½π‘π‘οΏ½
Therefore,
ππ(π‘π‘) = β ππππ(π‘π‘)βππ=1 = π π ππππ οΏ½ 1
1βππ(π‘π‘)οΏ½ (6)
Equation (6) implies that
οΏ½ΜοΏ½π(π‘π‘) = βπ π οΏ½1 βππ(π‘π‘)οΏ½οΏ½ΜοΏ½π(π‘π‘)
It can be verified that οΏ½ΜοΏ½π(π‘π‘) increases with πΎπΎ and decreases with πΌπΌ. Thus, the number of firms
in an industry increases over time more as πΎπΎ increases or as πΌπΌ decreases.
The rate of convergence or time to reach a steady state depends on the relative magnitude
of πΏπΏ and ππ because
12
οΏ½ΜοΏ½π(π‘π‘)ππ(π‘π‘)
= οΏ½πΏπΏοΏ½(1 β πΌπΌ)πΎπΎπΎπΎ + πποΏ½πΏπΏβπΏπΏππβοΏ½(1βππ)πΎπΎπΎπΎ+πποΏ½π‘π‘
βπποΏ½(1 β πΌπΌ)πΎπΎπΎπΎ + πποΏ½πΏπΏβππππβοΏ½(1βππ)πΎπΎπΎπΎ+πποΏ½π‘π‘
οΏ½ ππβοΏ½(1βππ)πΎπΎπΎπΎ+πποΏ½π‘π‘
Since πΏπΏ = πΌπΌππ + (1 β πΌπΌ)πΎπΎπΎπΎ + ππ, the convergence speed depends on frictions πΌπΌ and πΎπΎ too.
4. Data
The main data in this study come from the Japanese Census of Manufacture (Kogyo-Tokei)
between 1980 and 2013, supplemented with 2012 Economic Census for Business Activity.1 The
Ministry of Economy, Industry and Trade has been gathering information about Japanese
establishments in the manufacturing sector every year. All establishments with four employees or
more located in Japan are subject to the Census of Manufacture, and they are required by law to
answer survey questions. The Census of Manufacture contain various pieces of information about
establishments such as name, location, the number of employees, value of shipment, wage payment,
fixed capital, and so forth.
For this study, product information is the most important piece of information from the
Census of Manufacture. Each establishment is required to report a name of each product and a
value of each product it produced and shipped every year. Products in the Census of Manufacture
are classified into the six-digit level and the number of products at the six-digit level ranges from
about 2,200 to 2,300 categories over the period covered by this study. These six-digit level products
can be aggregated to the four-digit level, and there are about 450 to 550 products at the four-digit
level.2 For example, processed milk, milk beverage, butter, chees and ice cream are a different
1 The year 2012 is an Economic Census year, and the Census of Manufacture was conducted as a part of the 2012
Economic Census for Business Activity.
2 See the http://www.meti.go.jp/english/statistics/tyo/kougyo/index.html for the details of product classification.
13
product at the six-digit level, but all of them are classified into dairy products at the four-digit level.
In this study, an industry or a market is empirically defined at the four-digit level, and a
submarket is empirically defined at the six-digit level. By using these definitions, we empirically
identify the creation and destruction of a submarket in an industry, and investigate how the creation
and destruction of a submarket affect industry growth. A basic rationale for these definitions is that
many aspects of businesses such as production technology and knowhow and sales and distribution
systems are similar within such an industry to a larger extent than across such industries. One can
argue that these definitions would not be satisfactory ones because more disaggregated information
about products are available for some industries. The main purpose of this study, however, is to
examine whether some general pattern of industry growth emerges from a large data set containing
a number of industries. Therefore, some specific aspects of products and industries are sacrificed
in return for pursuing generalizability.
We also utilize data from the Basic Survey of Japanese Business Structures and Activities
(BSJBSA or Kitatsu, hereafter) to identify types of establishments empirically. This survey is
conducted by the Ministry of Economy, Industry and Trade every year, and collects information
from firms with at least 50 employees and 30 million yen paid-up capital. A particular importance
of this data set for the purpose of this study is to contain information about when and how firms
were established. In particular, firms answered whether they were established as a new firm or a
spinoff firm.
Firms in both the Census of Manufactures and the BSJBSA can be matched through name,
telephone number, and postal code matching. Once this matching is complete, a firm type can be
14
assigned to a firm, indicating whether it is a new firm, a spinoff firm or an incumbent firm in a
given year. Using the information in the Census of Manufacture about which firm owns which
establishment, establishments are classified to a new establishment, a spinoff establishment and an
incumbent establishment. One caveat of this matching procedure is that we can identify a type of
an establishment in the Census of Manufacture only if a firm that owns the establishment grew
large enough to be sampled in the BSJBSA. Some caution should be exercised when interpreting
empirical analyses about firm types.
5. Empirical Results
5. 1. Relationship between Submarket Creation and Positive Net Entry
I begin a regression analysis by examining how submarket creation and destruction affect a positive
net entry of establishments. In doing so, first, the Japanese manufacturing census data are divided
into two periods; (i) 1980 to 1985 and (ii) 1986 to 2013.3 Then, I empirically ask how submarket
creation and destruction between1980 and 1985 affect the length of positive entry period
between1986 and 2013.
A unit of analysis in this regression is an industry defined by the four-digit level product
category. Submarket creation is identified in the data by the creation of a new product at a six-digit
level category. This way of identifying submarket creation is unsatisfactory because there is a gap
in time between an actual submarket creation and statistical agencyβs recognition of a submarket
3 I tried several ways of dividing periods and got qualitatively the same results.
15
creation.4 Nonetheless, this gap will not be critical since this study focuses on a period of rapid
entry in the industry life-cycle. A submarket is identified to be destructed when no single
establishment produces the product of the submarket any longer or the submarket disappears from
the product classification.
A basic estimating equation of this regression analysis is given by
π¦π¦ππ = π½π½0 + π½π½1πΆπΆπππππππ‘π‘πΆπΆπΆπΆππππ + π½π½2π·π·πππππ‘π‘πππ·π·πππ‘π‘πΆπΆπΆπΆππππ + π½π½3πΆπΆπππππππ‘π‘πΆπΆπΆπΆππππ Γ π·π·πππππ‘π‘πππ·π·πππ‘π‘πΆπΆπΆπΆππππ + π½π½4π₯π₯ππ + ππππ
where π¦π¦ is the maximum periods of consecutive positive entry between 1986 and 2013,
πΆπΆπππππππ‘π‘πΆπΆπΆπΆππ is a dummy variable that takes 1 when at least one submarket in industry i is created
between 1980 and 1985. The control variables x includes total value of shipments of an industry,
and the number of submarkets within an industry.
Table 1 reports estimations results from OLS and hazard regressions. In the specification
(I), industries are classified into three categories; (i) an industry that created at least one submarket,
(ii) an industry that destructed at least one submarket, and (iii) an industry that neither created nor
destructed any submarket (the base category). According to the estimation result from this
specification, the coefficient on the submarket creation dummy is positive and statistically
significant at the 5 percent level whereas the coefficient on the submarket destruction dummy is
not statistically significant at the conventional significance levels. Although this result indicates
that submarket creation is likely to extend the period of positive net entry, this specification masks
an important fact about an impact of submarket creation and destruction on industry dynamics. In
4 As a general rule, the statistical agency reviews and modifies the current product classification about every five
years and create a new product category when a value of shipment of the product becomes a non-negregible share of
an induustry total.
16
the specification (II), the interaction term of submarket creation and submarket destruction is added
so that industries are now classified into four categories; (i) an industry that only created a
submarket, (ii) an industry that only destructed a submarket, (iii) an industry that both created and
destructed a submarket, and (iv) an industry that neither created nor destructed a submarket (the
base category).
The OLS estimation result from the specification II shows that, while the coefficient on
the market creation dummy is not statistically significant, the coefficient on the interaction term is
positive and statistically significant at the 10 percent significance level. The estimation result
suggests that the length of positive net entry becomes longer, relative to the base category, by about
two years when an industry both created and destructed a submarket. A hazard regression is also
conducted to address the truncation issue. The specifications (III) and (IV) present qualitatively the
same results as the OLS results. Since the hazard ratio of submarket creation and destructions is
less than the unity, a period of positive net entry is less likely to end when an industry both created
and destructed a submarket.
[Table 1 is about here]
A similar regression exercise is done by using entry rates as the dependent variable, instead
of using the maximum periods of positive entry. Table 2 reports estimation results from this exercise.
Specification (I) of Table 2 shows that the average entry rate during the period of 1986 to 2013
becomes higher in the case of submarket creation and destruction than the other cases. We see a
similar pattern in the specifications (II) and (IV), though the estimation result in the specification
17
(III) does not fit into this pattern.
[Table 2 is about here]
Overall, the estimation results in Tables 1 and 2 indicate that establishments are
encouraged to enter an industry when that industry both creates and destructs a submarket. This
main result can be also interpreted in the context of industry lifecycle. That is, the creation and
destruction of a submarket allow an industry to continue to attract new entrants so that the timing
of a shakeout is delayed. Thus, an industry tends to grow in terms of the number of establishments
when the industry evolves through the creation and destruction of submarkets.
5. 2. Who creates and destructs a submarket?
Since the previous section reveals the importance of submarket creation and destruction, it is
worthwhile to investigate what type of establishment is more likely to create or destruct a
submarket in a given industry. In particular, it is interesting to examine whether it is an
establishment owned by a startup firm, a spinoff firm or an incumbent firm that creates or destructs
a submarket. That said, the nature of the Japanese manufacturing census data does not permit us to
identify an establishment or a firm that has actually created or destructed a submarket. To see some
aspect of industry dynamics through this lens, the empirical question mentioned above is
reformulated as a question of what type of establishment is more likely to be engaged in a newly
created submarket or in a submarket that is about to be destructed.
18
Unlike the empirical analysis in the previous section, the first analysis of this section uses
an establishment as a unit of analysis and focuses on industries that created or destructed at least
one submarket. More specifically, I estimate the probability of being a newly created (destructed)
submarket:
ππππ(π·π·πππ‘π‘ = 1) = ππ(π½π½0 + π½π½1πππππΆπΆπππΆπΆπππππππ‘π‘ + π½π½2πππ‘π‘πππππ‘π‘π·π·πππππ‘π‘ + π½π½3π΄π΄πππππ·π·πΆπΆπππΆπΆπ‘π‘πΆπΆπΆπΆπππππ‘π‘ + π½π½4π₯π₯πππ‘π‘)
where π·π·πππ‘π‘ is a dummy variable that takes 1 if establishment i is in a newly created market in a
given industry at time t. The independent variables include establishment types and controls such
as total value of shipments, the number of products shipped and industry dummies. The variable
πππππΆπΆπππΆπΆπππππππ‘π‘ is equal to 1 if an establishment is owned by a firm that were legally separated as a
spinoff firm from a parent firm within the past five years from year t. The variable πππ‘π‘πππππ‘π‘π·π·πππππ‘π‘ is
equal to 1 if an establishment belongs to a newly started firm within the window of the past five
years from year t. The variable π΄π΄πππππ·π·πΆπΆπππΆπΆπ‘π‘πΆπΆπΆπΆπππππ‘π‘ is a dummy variable indicating whether an
establishment belongs to a firm that was created through acquisition process within the past five
years from year t. The baseline category is an establishment belonging to a firm operating in an
industry for more than 5 years.
Estimation results are presented in Table 3. In the Specification (I), all establishments
belonging to new firms are bunched together to see a role of new firms in a newly created
submarket. The coefficient on the new firm dummy is positive and statistically significant at the 1
percent significance level. This indicates that establishments belonging to new firms are more
likely to enter a newly created submarket than establishments belonging to incumbent firms. In the
Specification (II), heterogeneity of new firms are accounted for. According to the estimation result
19
from the specification (II), while the coefficient on the acquisition dummy is not statistically
significant, the coefficients on the spinoff dummy and startup dummy are positive and statistically
significant at the 10 percent and the 1 percent significance levels, respectively.
Turning our attention to a submarket that is about to be destructed, the estimation results
in the specifications (III) and (IV) pose a mirror image of the estimation results about a newly
created submarket. Startup firms and spinoff firms are more likely than incumbent firms to avoid
a submarket that is going to be destructed whereas acquisition firms are more likely to enter that
submarket.
The estimation results in Table 3 imply that firm age (i.e., young firms versus incumbent
firms) does not matter so much for active involvement in a newly created submarket or in a
destructed submarket, but different types of new firms play different roles in creation and
destruction of submarkets of a given industry.
[Table 3 is about here]
Next, I examine a similar issue from a slightly different viewpoint by focusing on the
distribution of firm types within an industry. More specifically, I investigate how the probability of
submarket creation or destruction in a given industry at time t+1 is affected by the existence of
particular firm types at time t. That is, the estimating equation is given by
ππππ(π·π·πππ‘π‘+1 = 1) = ππ(π½π½0 + π½π½1πππππΆπΆπππΆπΆπππππππ‘π‘ + π½π½2πππ‘π‘πππππ‘π‘π·π·πππππ‘π‘ + π½π½3π΄π΄πππππ·π·πΆπΆπππΆπΆπ‘π‘πΆπΆπΆπΆπππππ‘π‘ + π½π½4π₯π₯πππ‘π‘)
where π·π·πππ‘π‘+1 is equal to 1 if industry i creates a newly submarket at time t+1 (or destructed a
20
submarket), and firm types variables take on 1 if each type exists in industry i at time t. Note that
a unit of analysis switches back to an industry in this regression analysis.
Table 4 presents estimation results about such probabilities. According to Table 4, we can
see that a submarket is more likely to be created or destructed when there are startup firms in an
industry, but existence of spinoff firms and acquisition firms have no impact on such probabilities.
This appears to suggest that startup firms are critical for submarket creation and destruction,
although cautions should be exercised to interpret this result. As mentioned in Section 4, data
matching process is far from perfect. In particular, the matching process tends to pick up successful
startups and spinoffs and this may lead to a biased result.
[Table 4 is about here]
5. 3. Reallocation of unrealized opportunity with frictions
In this section, I use the model outlined in Section 3 to estimate frictions to pursuit of business
opportunities by incumbent firms (i.e., the parameter πΌπΌ ) as well as reallocation of unrealized
opportunities between incumbent firms and spinoff firms (i.e., the parameter πΎπΎ ). Once the
estimates of such frictions are obtained, I empirically examine how these frictions shape an industry
dynamics by influencing a net entry rate of the industry.
In section 3, the optimal conditions of an incumbent firm are derived as
ππβ²(ππ) = πΌπΌπ£π£
and
21
ππ = [ππ β πΌπΌππ β (1 β πΌπΌ)]π£π£ + ππ(ππ)
To proceed further, I assume that the cost function is quadratic with respect to ππ. That is,
ππ(ππ) = ππ0 + 12ππ1ππ2 . Under this assumption, these two optimal conditions for incumbent firms
yield
πππππππππππππππππππ‘π‘ = ππ0 + [ππ β (1 β πΌπΌ)]π£π£ β 12ππ2
ππ1π£π£2 (7)
Equation (7) forms a basis for estimating πΌπΌ. Given the interest rate ππ, the parameter πΌπΌ can be
estimated from the regression of πππππππππππππππππππ‘π‘ on π£π£. Similarly, for incumbent firms, we have
πππ π π π πππππ π π π π π = ππ0 + [ππ β πΌπΌ(1 β πΎπΎ)]π£π£ β 12πΎπΎ2(1βππ)2
ππ1π£π£2 (8)
Equation (8) implies that the parameter πΎπΎ can be estimated from the regression of πππ π π π πππππ π π π π π on π£π£
once we obtain an estimate of πΌπΌ and ππ.
Data on (ππ, π£π£) can be calculated from profit data of each establishment. For this analysis,
ππ is the current profit of an establishment calculated from the Census of Manufacture, and π£π£ is a
sum of discounted future profits. Since the panel data are only available after 1986, I calculate the
average current profits of each establishment between 1986 and 1990 and use this average as the
current profit ππ. Regarding the calculation of π£π£, I use profits of each establishment between 1991
and 2013 as a stream of future profits. Then, ππ is regressed on π£π£ and π£π£2 separately for a set of
incumbent establishments and a set of spinoff establishments in a given industry to estimate πΌπΌ and
πΎπΎ. To calculate πΌπΌ and πΎπΎ, the interest rate ππ is set to 0.05.
Table 5 presents the distributions of estimates of πΌπΌ and πΎπΎ across industries. The mean
of estimates of πΌπΌ is about 0.992, and it suggests that there is some friction to the pursuit of
22
business opportunities by incumbent firms. We can also observe that there is some variation in the
estimates of πΌπΌ and πΎπΎ across industries. Such variations are now used to examine to what extent
these frictions are critical for variation in the number of net entry across industries.
[Table 5 is about here]
Table 6 shows estimation results about a relationship between frictions and entry rates.
According to Table 6, the coefficient on πΌπΌ is negative and is not statistically significant at the
conventional significance levels whereas the coefficient on πΎπΎ is statistically significant at the 5
percent or the 10 percent significance level.
Notice that a larger value of πΌπΌ indicates a lesser degree of friction to the pursuit of
business opportunities by incumbent firms. Although the coefficient on πΌπΌ is not statistically
significant, the negative sign of the coefficient is in line with the modelβs prediction. When πΌπΌ is
high, incumbent firms realize many business opportunities in a given industry so that there is a
little room for startup firm and spinoff firms to enter that industry.
A larger value of πΎπΎ means a lesser degree of friction to reallocation of unrealized
opportunities from incumbent firms to spinoff firms. In other words, spinoff firms can pursue
unrealized opportunities of incumbent firms more often as a value of πΎπΎ becomes larger. The
positive sign of the coefficient on πΎπΎ can be interpreted as indicating that new entry is encouraged
when unrealized business opportunities are reallocated smoothly to spinoff firms from incumbent
firms. When product innovation is cumulative process in that one innovation becomes a foundation
23
for next innovations (Aghion et al., 2008), subsequent innovations may not come to existence if
some frictions do not allow some firms to pursue unrealized innovation opportunities of other firms.
In light of this insight, the estimation results in Table 6 seem to imply that this cumulative nature
of business and innovation opportunities generates self-enhancing process of new entry in the sense
that reallocation of unrealized opportunities from incumbent firms to spinoff firms sparks
subsequent entries.
[Table 6 is about here]
6. Conclusion
This study used the large-scale data of the Japanese Census of Manufacture between 1980 and 2013
to empirically investigate the process of Schumpeterian industry growth, its generalizability and
the role of reallocation of unrealized opportunities in industry dynamics. The empirical analyses of
this study showed that the creation and destruction of products have a positive impact on the length
of entry periods and entry rates, that startup firms and spinoff firms are more likely than incumbent
firms to be involved in a newly created submarket and that an industry continues to attract new
entrants when reallocation of unrealized opportunities among different firm types functions well
without serious frictions.
Several caveats need to be recognized in order to interpret the main findings of this study
appropriately. First, this study does not intend to establish and claim any causal relationship. All
the findings of this study should be interpreted as indicating some association among the research
24
variables of this study. Second, some insights from the empirical analyses of this study rely on
indirect evidence as well as strong assumptions about functional forms. For example, the friction
parameters were estimated by assuming that the cost function was quadratic with respect to
innovation intensity. Some generalization about the functional form needs to be investigated since
an alternative specification may change the estimation results qualitatively. Finally, the sample used
in the empirical analyses biases towards establishments belonging to successful firms. This bias
may affect all types of firms β startup, spinoff and incumbent firms β in the same direction, but it
is hard to detect and infer the direction of this bias. Although these caveats raise some concern, this
study presented interesting relationships between innovation, firm heterogeneity and industry
dynamics.
25
References
P. Aghion, M. Dewatripont, and J.C. Stein. βAcademic Freedom, Private-Sector Focus, and the
Process of Innovation.β Rand Journal of Economics, Vol.39, 2008, pp.617-635.
P. Aghion and P. Howitt. βA Model of Growth Through Creative Destruction.β Econometrica,
Vol.60, 1992, pp. 323-351.
P. Aghion and J. Tirole. βFormal and Real Authority in Organization.β Journal of Political
Economy, Vol.105, 1997, pp.1-29.
A. Barbarino and B. Jovanovic. βShakeouts and Market Crushes.β International Economic Review,
Vol.48, 2007, pp.385-420.
A. Collard-Wexler and J. De Loecker. βReallocation and Technology: Evidence from the U.S. Steel
Industry.β American Economic Review, Vol.105, 2015, pp. 131-171.
D. Filson. βThe Nature and Effects of Technological Change over the Industry Life Cycle.β Review
of Economic Dynamics, Vol.4, 2001, pp.460-494.
P. Geroski. βWhat Do We Know about Entry?β International Journal of Industrial Organization,
Vol.13, 1995, pp. 421-440.
R. Golman and S. Klepper. βSpinoffs and Clustering.β Rand Journal of Economics, Vol.47, 2016,
pp.314-365.
M. Gort and S. Klepper. βTome Paths in the Diffusion in the Product Innovations.β Economic
Journal, Vol.92, 1982, pp.630-653.
C. Hsieh and P.J. Klenow. βMisallocation and Manufacturing TFP in China and India.β Quarterly
Journal of Economics, Vol. 124, 2009, pp. 1403-1448.
26
H. Itoh and O. Hayashida. βBunkenka to Kengen Ijo.β Nihon Keizai Kenkyu, No. 34, 1997, pp. 89-
117.
B. Jovanovic and G. MacDonald. βThe Lifecycle of a Competitive Industry.β Journal of Political
Economy, Vol.102, 1994, pp.322-347.
T.J. Klette and S. Kortum. βInnovating Firms and Aggregate Innovation.β Journal of Political
Economy, Vol.112, 2004, pp.986-1018.
S. Klepper Experimental Capitalism, 2016, Princeton University Press.
S. Klepper and E. Graddy. βThe Evolution of New Industries and the Determinants of Market
Structure.β Rand Journal of Economics, Vol. 21, 1990, pp. 27-44.
S. Klepper and K. Simons. βThe Making of Oligopoly: Firm Survival and Technical Cange in the
Evolution of the US Tire Industry.β Journal of Political Economy, Vol.51, 2000 pp.728-760.
S. Klepper and S. Sleeper. βEntry by Spin-offs.β Management Science, Vol.51, 2005, pp.1291-1306.
S. Klepper and P. Thompson. βSubmarkets and the Evolution of Market Structure.β Rand Journal
of Economics, Vol.37, 006, pp.861-886.
R. Lentz and D.T. Mortensen. βAn Empirical Model of Growth through Product Innovation.β
Econometrica, Vol. 76, 2008, pp. 1317-1373.
M. Otsubo. The Corporate Restructurings of Japanese Firms. 2005, Chuou Keizaisya.
J. Schumpeter. Capitalism, Socialism, and Democracy. 1942, Harper & Brothers.
J.P. Tang. βA Tale of Two Sics: Japanese and American Industrialisation in Historical Perspective.β
Australian Economic History Review, Vol. 56, 2016, pp. 174-197.
27
Tables
Table 1: Positive Net Entry Periods and Submarkets
Dependent variable: Positive Entry Periods
OLS
Hazard Ratio
I
II
III
IV
Submarket Creation 0.7455 ** 0.5041
1.364
1.28946
(0.3685)
(0.3899)
(2.819)
(1.65473)
Submarket Destruction 0.4424
0.1973
1.365
1.27897
(0.3512)
(0.3744)
(1.909)
(1.79702)
Creation & Destruction
1.9854 *
0.00004 **
(1.0728)
(0.000031)
Value of Shipments 0.0004 ** 0.0004 **
0.997
0.99739
(0.0002)
(0.0002)
(0.002)
(0.00213)
Number of Products -0.0966 *** -0.0853 **
1.247 *** 1.25210 **
(0.0344)
(0.0349)
(0.127)
(0.12917)
No. of observations 440 440 440 440
Note: ***, **, and * indicate that a coefficient is statistically significant at 1 percent, 5 percent and 10 percent significance levels,
respectively.
28
Table 2: Net Entry Rates and Submarkets
Dependent variable: Entry Rates
I
(1986 - 2013)
II
(1986 β 1990)
III
(1996 β 2000)
IV
(2006 β 2010)
Submarket Creation -0.003440
0.007442 * 0.0096742
0.000221
(0.006166)
(0.004039) (0.0061287)
(0.007510)
Submarket Destruction -0.000089 -0.002062
-0.0027783 0.003000
(0.005966)
(0.003908)
(0.0059305)
(0.007267)
Creation & Destruction 0.030530 * 0.027452 ** 0.0136738
0.045175 **
(0.017152)
(0.011236)
(0.0170497)
(0.020892)
Value of Shipments 0.000003
0.000003
2.50E-06
0.000003
(0.000003)
(0.000002) (0.000002)
(0.000003)
Number of Products -0.001473 *** -0.000961 *** -0.0017621 *** -0.000525
(0.000548)
(0.000359)
(0.0005445)
(0.000667)
No. of observations 440 440 440 440
Note: ***, **, and * indicate that a coefficient is statistically significant at 1 percent, 5 percent and 10 percent significance levels,
respectively.
29
Table 3: Net Entry Rates and Submarkets
Logit model
Marginal effects
In Newly Created
Submarket
In Newly Destructed
Submarket
I
II
III
IV
New firm
dummy 0.0249 ***
-0.0182 ***
(0.0051) (0.0048)
Spinoff dummy 0.0260 * -0.0279 *
(0.0155) (0.0159)
Acquisition
dummy 0.0036
0.0445 ***
(0.0151) (0.0095)
Startup dummy 0.0280 *** -0.0357 ***
(0.0057) (0.0060)
Observations
672,543
672,543
823,060
823,060
Year Dummy Yes
Yes
Yes
Yes
Controls Yes
Yes
Yes
Yes
Note: ***, **, and * indicate that a coefficient is statistically significant at 1 percent, 5 percent and 10 percent significance levels,
respectively.
30
Table 4: Firm Types. Submarket Creation and Destruction
Mlogit model
Marginal effects
I
Creation
II
Destruction
Spinoff dummy -0.0013 0.0235
(0.0190)
(0.0207)
Acquisition
dummy 0.0223
0.0117
(0.0182) (0.0205)
Startup dummy 0.0674 *** -0.0616 ***
(0.0220) (0.0201)
Observations 1,481 1,481
Year Dummy Yes
Yes
Controls Yes Yes
Note: ***, **, and * indicate that a coefficient is statistically significant at 1 percent, 5 percent and 10 percent significance levels,
respectively.
31
Table 5: Estimation of Friction Parameters πΌπΌ and πΎπΎ
Mean Std. Dev. 10th 25th 50th 75th 90th
alpha 0.9916 0.0288 0.9599 0.9744 0.9887 1.0044 1.0303
gamma 1.0383 0.2081 0.9279 0.9514 0.9979 1.0637 1.1540
Note: ***, **, and * indicate that a coefficient is statistically significant at 1 percent, 5 percent and 10 percent significance levels,
respectively.
Table 6: Entry Rates and Frictions
Dependent variable: Entry Rates
I II III IV
(1991 β 2013) (1991 β 1995) (1996 β 2000) (2001 β 2005)
alpha -0.0648 -0.2025 -0.1050 -0.2052
(0.1133) (0.1373) (0.0868) (0.2272)
gamma 0.0114 ** 0.0172 ** 0.0260 *** 0.0260 ***
(0.0050) (0.0078) (0.0054) (0.0083)
No. of
observations 97 97
97 97
Note: ***, **, and * indicate that a coefficient is statistically significant at 1 percent, 5 percent and 10 percent significance levels,
respectively.